Throughput measurement tools

Blast Manual

The tool consists of four executables BlastI, BlastC, BlastS and snjrt20. The latter contains the dll libraries needed for the first three, but will never be addressed explicitly. For people who do not want to download the binaries, the Graphical User Interfaces of the three programs can be viewed through these links: BlastIGUI, BlastSGUI, BlastCGUI. Note that the throughputs displayed on the windows are bus throughputs and have nothing to do with wireless LAN.
BlastI.exe is the executable that initializes and starts a test. For a test, one BlastI.exe has to be running on one of the clients. The parameters block size, number of blocks and number of clients to participate need to be set here. The function of BlastI is to broadcast to all clients (where a BlastC is running). This broadcast contains the test parameters and serves as a start signal to all clients, who have a client ID lower than number of clients.
BlastC.exe is the client running on the mobile stations. On the 'go' sign from BlastI each client starts to send data to its designated server. The server host and port can be set, as well as a client ID. This client ID serves to separate the various clients. One can set a number of clients parameter on BlastI. Only clients with a client ID below 'number of clients' will start sending on arrival of the broadcast. It is thus advised to enumerate the clients 1,2,3,4,5.... Another possibility is to cluster the clients into groups by giving all blue ones client ID 1, the green clients get ID 2, etc. The throughputs attained in each session are logged in a file (settable name, default Blast.out in the directory BlastC is running in) in the following format:

rank/ID, number of blocks, block size, transmission time, and throughput in Mbit/sec.

Finally, a BlastServer is needed to receive the data sent by all clients. One running BlastS.exe is enough, as different threads can listen at different port numbers. These do have to match the port numbers set at the BlastClients (1 to 1).
In our setup the BlastInitializer and BlastServer are running on the wired PC. On each client one BlastClient is running.
 

Effects on testing

Compared to the previous test this upgraded version is clearly much easier to use, but it also provides better measurements. The uncertainties introduced by the unequal startup times and the load on the wired PC are circumvented by this new tool. After the BlastInitializer sends the broadcast all clients start sending data within one second. With measurement times of roughly one minute, errors due to startup are negligible. Measurement durations cannot be chosen much smaller than one minute, as statistical errors are then starting to play a role.
Further, the clients are now running on the mobile stations. The data rate in this configuration is certainly not limited by the performance of the wired PC. The only load on the wired PC is the BlastInitializer, giving one broadcast before data transmission starts, and the BlastServer, whose function is to receive and flush the data (2 Mbit/second maximum).